`torchtriton` on `conda` is not up-to-date, leading to CPU memory leak

[konstantinjdobler]

🐛 Describe the bug

#96937 describes a memory leak when using torch.compile due to a bug in triton. This bug was fixed and triton was updated -- but only on pip! The current version of triton on conda that is installed with pytorch (pytorch/torchtriton) has not been updated. Installing the latest torchtriton via pytorch-nightly fixes the memory leak.

To fix: please update pytorch/torchtriton to the latest version. The updated triton version that contains the fix did not actually change the SemVer version (it's still 2.0.0.post1 vs. 2.0.0 previously), which might be why the conda version was not auto-updated.

Versions

PyTorch version: 2.0.0
Is debug build: False
CUDA used to build PyTorch: 11.8
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.1 LTS (x86_64)
GCC version: (Ubuntu 11.3.0-1ubuntu1~22.04) 11.3.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.35

Python version: 3.10.10 | packaged by conda-forge | (main, Mar 24 2023, 20:08:06) [GCC 11.3.0] (64-bit runtime)
Python platform: Linux-5.15.0-56-generic-x86_64-with-glibc2.35
Is CUDA available: False
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: 
GPU 0: NVIDIA RTX A6000
GPU 1: NVIDIA RTX A6000
GPU 2: NVIDIA RTX A6000
GPU 3: NVIDIA RTX A6000
GPU 4: NVIDIA RTX A6000
GPU 5: NVIDIA RTX A6000
GPU 6: NVIDIA RTX A6000
GPU 7: NVIDIA RTX A6000

Nvidia driver version: 515.86.01
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture:                    x86_64
CPU op-mode(s):                  32-bit, 64-bit
Address sizes:                   46 bits physical, 48 bits virtual
Byte Order:                      Little Endian
CPU(s):                          48
On-line CPU(s) list:             0-47
Vendor ID:                       GenuineIntel
Model name:                      Intel(R) Xeon(R) Silver 4214 CPU @ 2.20GHz
CPU family:                      6
Model:                           85
Thread(s) per core:              2
Core(s) per socket:              12
Socket(s):                       2
Stepping:                        7
CPU max MHz:                     3200.0000
CPU min MHz:                     1000.0000
BogoMIPS:                        4400.00
Flags:                           fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req pku ospke avx512_vnni md_clear flush_l1d arch_capabilities
Virtualization:                  VT-x
L1d cache:                       768 KiB (24 instances)
L1i cache:                       768 KiB (24 instances)
L2 cache:                        24 MiB (24 instances)
L3 cache:                        33 MiB (2 instances)
NUMA node(s):                    2
NUMA node0 CPU(s):               0-11,24-35
NUMA node1 CPU(s):               12-23,36-47
Vulnerability Itlb multihit:     KVM: Mitigation: VMX disabled
Vulnerability L1tf:              Not affected
Vulnerability Mds:               Not affected
Vulnerability Meltdown:          Not affected
Vulnerability Mmio stale data:   Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Retbleed:          Mitigation; Enhanced IBRS
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1:        Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:        Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Vulnerability Srbds:             Not affected
Vulnerability Tsx async abort:   Mitigation; TSX disabled

Versions of relevant libraries:
[pip3] flake8==6.0.0
[pip3] mypy-extensions==0.4.3
[pip3] numpy==1.23.5
[pip3] pytorch-lightning==2.0.2
[pip3] torch==2.0.0
[pip3] torchmetrics==0.11.4
[pip3] triton==2.0.0
[conda] blas                      2.116                       mkl    conda-forge
[conda] blas-devel                3.9.0            16_linux64_mkl    conda-forge
[conda] libblas                   3.9.0            16_linux64_mkl    conda-forge
[conda] libcblas                  3.9.0            16_linux64_mkl    conda-forge
[conda] liblapack                 3.9.0            16_linux64_mkl    conda-forge
[conda] liblapacke                3.9.0            16_linux64_mkl    conda-forge
[conda] mkl                       2022.1.0           h84fe81f_915    conda-forge
[conda] mkl-devel                 2022.1.0           ha770c72_916    conda-forge
[conda] mkl-include               2022.1.0           h84fe81f_915    conda-forge
[conda] numpy                     1.23.5          py310h53a5b5f_0    conda-forge
[conda] pytorch                   2.0.0           py3.10_cuda11.8_cudnn8.7.0_0    pytorch
[conda] pytorch-cuda              11.8                 h7e8668a_3    pytorch
[conda] pytorch-lightning         2.0.2              pyhd8ed1ab_0    conda-forge
[conda] pytorch-mutex             1.0                        cuda    pytorch
[conda] torchmetrics              0.11.4             pyhd8ed1ab_0    conda-forge
[conda] torchtriton               2.0.0                     py310    pytorch

cc @ezyang @gchanan @zou3519 @seemethere @malfet @msaroufim @wconstab @ngimel @bdhirsh @anijain2305 @soumith

[Chillee]

I'm not sure if this is working as intended? cc: @malfet

[konstantinjdobler]

Is there anything I can do to help with this? @malfet @Chillee

The last update of torchtriton in the pytorch channel was on the 4th of March 2023 but the commit to fix the memory leak in #96937 was pushed to openai/triton after that (17th of March).

[msaroufim]

We've had issues with updates to the conda index. This might be related cc @atalman

[MihaiBabiac]

Is there any progress on this? With conda being the recommended way of installing pytorch, I'd say it's a pretty significant issue that compiled models have a memory leak.

On a side note, triton pushing a bugfix without bumping the version doesn't make much sense from a user's point of view. Conda and pip think that nothing has changed so the bugfix doesn't get pulled automatically. Currently I have to remove triton from the conda env and manually install it with pip, which is quite hacky:

micromamba remove -f torchtriton
pip3 install triton --index-url https://download.pytorch.org/whl/cu118

[konstantinjdobler]

With the latest release of torchtriton on conda, this has been resolved for pytorch>=2.1.0.